Owlish is live: AI customer support, grounded in your knowledge

I started Owlish because I was tired of cleaning up after chatbots.

Every team I worked with had the same story. Someone bought a chatbot. It promised refunds that didn’t exist, made up shipping policies, hallucinated discount codes. Tickets came in louder than before. The “AI” became a layer of risk on top of a support team that was already busy.

So we built the thing we wished existed: an AI agent that answers from your sources, cites the page it pulled from, and hands the conversation to a human the moment it isn’t sure. Today, that’s live at owlish.bot.

What Owlish actually is

Owlish is a no-code AI agent platform for customer support. You give it your knowledge — websites, PDFs, FAQs — and you get an agent that runs across the channels your customers already use: a web widget, Slack, Teams, Discord, and email. One agent, many surfaces, one inbox to manage them all.

The pitch is short, but three things make it different in practice:

1. Every answer cites its source

If the agent can’t ground its answer in your knowledge base, it doesn’t guess. It says it doesn’t know and offers to hand the conversation over. Every grounded answer comes back with a clickable link to the page or document it used, so the customer can verify and your operator can spot-check.

This isn’t a premium tier. It’s the only way Owlish works.

2. Operators are first-class, not an afterthought

The Helpdesk inbox shows every session — from the widget, from Slack, from email — in one place, with the full transcript, citations, and a one-click handoff button. Whisper mode lets a human nudge the agent’s next reply without taking the conversation over. CSAT, deflection rate, and gap topics show up in the same view, so you know what the agent is actually doing.

If your “AI agent” can’t hand off cleanly, it’s not saving support time — it’s hiding it.

3. Deploy once, run everywhere

A single agent powers the widget, Slack, Teams, Discord, and email. Update the knowledge base in one place; every channel updates. There’s no “agent for Slack” and “agent for the website” to keep in sync.

What ships in v1

The launch lineup, in plain language:

• Knowledge ingestion. Crawl a website, drop in PDFs, paste FAQ pairs. We chunk, embed, and cite the original page in every answer.
• Multi-channel deployment. Web widget, Slack, Teams, Discord, and email out of the box. Add a channel; the same agent picks up.
• Agent customization. Pick a model (Gemini, Claude, GPT), write a persona, set the welcome message, define what to do when the agent doesn’t know.
• Helpdesk inbox. Shared inbox, transcript view with citations, whisper mode, one-click human handoff, role-based access for your team.
• Activity dashboards. Deflection rate, gap topics, CSAT, top sources. The numbers a support lead actually checks on a Monday morning.
• Workspace isolation by default. Your data is yours. Encrypted at rest and in transit. Never used to train shared models. Exportable. Deletable on demand. GDPR & CCPA aligned out of the box.

You can see the full feature list and what’s coming next on the roadmap, and the deeper “what it is” tour is over here.

Pricing in one paragraph

Start on Free with one no-card workspace. Paid plans start at Starter and scale up for teams that need more capacity, channels, and retention. We meter on chat sessions and knowledge capacity, not AI credits. Full breakdown on the pricing page.

Why now

Two things changed.

The model side caught up: grounded answers with citations are now fast enough to feel like a real chat, not a research tool. The operator side caught up too: support teams have spent the last few years moving to shared inboxes, and a chat agent that can’t cleanly hand off into one of those is now a square peg.

We think the next wave of support tooling won’t be “chatbot bolted onto a help desk.” It’ll be one platform where the AI agent and the human team share the same inbox, the same knowledge, and the same responsibility for getting the customer the right answer. That’s what we’re building.

What I want from you

If you run a support team — five operators or fifty — I’d love for you to break Owlish.

• Start free and point it at your real help center.
• Try to make it hallucinate. Tell me where it cracks.
• Then email me directly or grab 15 minutes on Cal.com and tell me what’s missing.

We ship in public. The roadmap is online, product notes are public, and customer-specific planning stays private. If something is broken, I want to know about it before you do.

Welcome to Owlish.

— Mithun